Top-P Sampling
Domain: NLP Fundamentals
🎯 Motivation
Samples from the smallest set of tokens whose cumulative probability reaches P (nucleus sampling).
📋 Concrete Example
Top-P = 0.9 means take tokens in order of probability until they add up to 90%, then sample from that set.
🔗 Analogy
Like packing a suitcase - take the most important items first until you reach 90% full, then pick randomly from the rest.